Hugging Face's logo

totally-not-an-llm
/
EverythingLM-13b-V3-16k

Text Generation
Transformers
PyTorch
llama
text-generation-inference
Inference Endpoints
Model card Files Community
2
EverythingLM-13b-V3-16k / README.md
totally-not-an-llm's picture
totally-not-an-llm
Update README.md
d338140
preview code
|
raw
history blame
1.13 kB
metadata
license: llama2
datasets:
  - totally-not-an-llm/EverythingLM-data-V3

EverythingLM-13b-V3-16k

Introducing EverythingLM, a llama-2 based, general-purpose 13b model with 16k context thanks to LlongMa. The model is trained on the EverythingLM-V3 dataset, more info can be found on the dataset page.

The model is completely uncensored.

Despite being "uncensored", the base model might be resistant; you might have to prompt-engineer certain prompts.

Notable features:

  • Automatically triggered CoT reasoning.
  • Verbose and detailed replies.
  • Creative stories.
  • Good prompt understanding.

Differences from V2:

  • General all around improvements thanks to the new dataset. Check out the dataset for more info.

Prompt format (Alpaca-chat): 

USER: <prompt>
ASSISTANT:

Future plans:

  • Highest priority right now is V3.1 with more optimized training and iterative dataset improvements based on testing.

Note:

Through testing V2, I realized some alignment data had leaked in, causing the model to be less cooperative then intended. This model should do much better due to stricter filetering.